A conversation analysis framework using speech recognition and naïve bayes classification for construction process monitoring

Abstract

At a dynamic construction site, conversations convey vital information including construction activities, operation status, and task performance. Even though because of information security, recording the entire conversations of a construction project is currently somewhat restricted, establishing a framework to capture and analyze construction conversations would be a promising approach to enhance the utilization of new field information for construction progress monitoring and safety surveillance. The construction industry, however, has no proper method to deal with onsite conversations. To enhance construction process and safety monitoring that is crucial for construction project management, this paper proposes a new framework to acquire onsite conversations and analyze their significance and interrelationship. The proposed conversation analysis framework involves the integrated implementation of the speech recognition library and the Natural Language Processing Toolkit using the Naive Bayes classifier, which helps translate the conversations to a text script and classify them according to the distinct types of construction activities and operations. Using the conversation videos, this paper represents the translation and classification accuracy of construction relevant conversations. The web audio and text data related to three possible conversation topics at a construction site were collected and used to test the framework in this paper. The proposed framework reached 90.9% overall accuracy. This research is expected to help domain experts monitor construction work processes and make data-driven decisions based on analyzed onsite conversation data.